G. Antoniou et al. (Eds.): SETN 2006, LNAI 3955, pp.

105

115, 2006.
Springer-Verlag Berlin Heidelberg 2006
Improved Wind Power Forecasting Using a Combined
Neuro-fuzzy and Artificial Neural Network Model
Yiannis A. Katsigiannis
1
, Antonis G. Tsikalakis
2
,
Pavlos S. Georgilakis
1
, and Nikos D. Hatziargyriou
2

1
Department of Production Engineering and Management, Technical University of Crete,
University Campus, Kounoupidiana, Chania, Greece
{katsigiannis, pgeorg}@dpem.tuc.gr
2
School of Electrical and Computer Engineering, National Technical University of Athens,
Athens, Greece
{atsikal, nh}@power.ece.ntua.gr
Abstract. The intermittent nature of the wind creates significant uncertainty in
the operation of power systems with increased wind power penetration. Con-
siderable efforts have been made for the accurate prediction of the wind power
using either statistical or physical models. In this paper, a method based on Ar-
tificial Neural Network (ANN) is proposed in order to improve the predictions
of an existing neuro-fuzzy wind power forecasting model taking into account
the evaluation results from the use of this wind power forecasting tool. Thus, an
improved wind power forecasting is achieved and a better estimation of the
confidence interval of the proposed model is provided.
Keywords: Artificial neural networks, wind power forecasting, prediction error.
1 Introduction
Wind power is one of the dominant Renewable Energy Sources (RES) since, by the
end of 2004, over 47 GW have been installed worldwide, 34 GW of which in Europe
[1]. In Greece, the installed wind power capacity is 567 MW, 164.5 MW of which in
the autonomous power systems of Greek islands [2]. The intermittent nature of wind
power production forces the power systems operators maintaining significant percent-
age of spinning reserve to compensate for uncertainties in wind power product-ion.
Sometimes, especially in autonomous power systems with increased wind power
penetration, operators may even consider totally unreliable the wind power production
leading the system to operate with excessive spinning reserve and thus increasing its
operating cost.
In the past few years, there have been several studies on wind power forecasting.
The simplest method of all, more suitable for shorter prediction horizon, is the persis-
tence method, considering that the expected wind power production in the following
few hours will be the same as the current hour. The accuracy of the persistence
method is reduced as the prediction horizon is increased. Wind power forecasting
methods include models based on statistical methods as presented in [3] and methods
based on Artificial Neural Networks (ANN), e.g. Radial Basis Functions topology [4]
106 Y.A. Katsigiannis et al.
or adaptive Fuzzy-Neural networks [5, 6]. Some efforts have been also made with
time series and ARMA models, requiring however, transformation and standardiza-
tion, given the non-Gaussian nature of the hourly wind speed distribution and the non-
stationary nature of its daily evolution [7]. A more detailed literature overview of the
developed wind power forecasting tools is described in [8]. Some of these methods
use meteorological information, mainly wind speed, especially for longer period fore-
casts, provided by Numerical Weather Prediction (NWP) models like SKIRON and
Hirlam.
The impact of improved wind forecasting tool with actual data for the last 4
months of 2001 has shown that improvement of wind forecasting errors has signifi-
cant economic impact in the operation of the power system due to the reduction of
spinning reserve requirements [9]. The reduction in the operating cost is about 1.8%-
3.5% if a reliable forecast is used that allows the reduction of spinning reserve in the
50% of wind power production. The reduction in the operating cost is about 2.3%-
5.3% if a reliable forecast is used that allows the reduction of spinning reserve in the
20% of wind power production. Therefore, the more reliable the wind power forecast-
ing is, the more confident the operators of the power systems are for the wind power
production forecast and thus, the spinning reserve requirements can be further re-
duced, leading to the reduction of the power system operating cost.
The developers of wind power forecasting models provide their end-users with the
Mean Absolute Percentage Error (MAPE) index for their model expressed as a per-
centage of the installed wind power capacity. This index, however, does not give very
much information neither about the performance of the wind power forecasting tool
for different forecasting horizon nor about its performance for a variety of forecasted
wind power values. Some wind power forecasting tools also provide as output the
confidence interval of the wind power forecast based on the estimation of the weather
stability and other parameters having to do with the forecasting model itself [10].
Such information helps the operators to estimate the range of the expected wind
power production and thus the spinning reserve requirements to cope with the wind
power production uncertainty.
In this paper, a method is proposed based on ANN, in order to improve the per-
formance of an existing wind power forecasting tool. This method uses as inputs the
outputs of the wind power forecasting model and trains the ANN using the results
from the evaluation of the forecasting model. The output of the ANN is a new and
improved wind power forecast. Moreover, an 85% confidence interval is provided to
the operators for this improved wind power forecast.
The methodology followed to derive the improved wind power forecast is de-
scribed in detail in Section 2. This methodology is applied to the wind power forecast-
ing model developed within the MORE CARE framework [11, 12] that was executed
off-line to produce wind power forecasts for a period with available meteorological
data from SKIRON for the power system of Crete. Some information on the power
system of Crete is provided in Section 3 concerning mainly the wind power. Section 4
presents results from the application of the proposed methodology to the power sys-
tem of Crete evaluating the improved forecast obtained using as a criterion the change
in the 85% interval and the MAPE. Conclusions are drawn in Section 5.
Improved Wind Power Forecasting Using a Combined Neuro-fuzzy and ANN Model 107
2 Improved Wind Power Forecasting
In this paper, an existing neuro-fuzzy wind power forecasting tool, considered as a
black box, is combined with an ANN, whose general structure is shown in Fig. 1, in
order to improve the accuracy of the wind power forecast.

Inputs:
1-24 hour-ahead predictions
of total wind power
provided by the
neuro-fuzzy model
Indicator of current hour
Output:
Wind Power
of the
following
i-th hour
(i=1,,24)
ANN
24
1
1

Fig. 1. General structure of the combined ANN and Neural Fuzzy Network
The improved wind power forecasting methodology consists of the following 4
steps:
1. Creation of two independent Data Sets (DS) by off-line execution of the fore-
casting model,
2. Split of DS into Learning Set (LS) and Test Set (TS),
3. Creation and Training of the ANNs,
4. Evaluation of the ANN outputs and confidence interval derivation.
2.1 Preparation of the LS and TS
The DS for the ANN model is created as follows: The MORE CARE wind power
prediction tool was run off-line for the last 4 months of 2001 providing forecasts for
each hour at 24 hour steps. The next 24 hours forecasted values plus an indicator for
the current hour are used as inputs for the DS, which consists of 663 time-series in our
case study. This DS contains periods of various wind power production levels ranging
from very low to very high wind power production.
To ensure more reliable results and to avoid confidence intervals with values below
zero or above the wind power capacity, the DS is split into two major classes accord-
ing to the forecasted values: the first one, with half the data contains values of 0-10
MW (DS1) and the second one with the rest available predictions has prediction val-
ues of 10-67.35 MW (DS2).
Each DS is split into a LS and TS. In our case, 2/3 of the data in each DS was used
for training and 1/3 was used for the test. The TS data was used for estimating the
confidence interval of the existing forecasting tool. Thus, an objective comparison
with the same set of data can be performed.
2.2 Creation and Training of the ANNs
For each one of DS1 and DS2 and for each hour, an ANN has been developed, thus a
total number of 48 ANNs has been used.
After the training procedure, the neural network is able to learn (generalize) the in-
put-output relationship and thus to predict the wind power to any input vector outside
the training set.However, good generalization depends on the network structure. In
108 Y.A. Katsigiannis et al.
particular, small size networks are not able to approximate complicated input-output
relationships. On the other hand, recent studies on learning versus generalization
network capabilities including the VC dimension [13] indicate that an unnecessarily
large network size heavily deteriorating generalization. In our approach, we adopt a
back-propagation variant [14] in a constructive framework [15], which begins with a
small size network and subsequently adds neurons to improve the network perform-
ance. A validation data set has been also used during training to control learning with
respect to the generalization ability of the network.
In Table 1, the results of different extensively studied ANN architectures for a va-
riety of hour-ahead predictions are presented. The selected architecture is the one with
the minimum MAPE during the whole prediction horizon. In the specific study, the
optimal ANN structure for both classes was the one consisting of 3 hidden layers of
13 neurons each, namely 25-13-13-13-1. In Fig. 2, the performance of the selected
ANN architecture for different number of epochs is examined as far as MAPE is con-
cerned. According to this figure, the optimal number of epochs was 15.
Table 1. MAPE of TS in the 10-67.35 MW class for different ANN architectures
MAPE of 10-67.35 MW class ANN architecture
1 hour ahead
prediction
12 hour ahead
prediction
24 hour ahead
prediction
25-13-1
25-25-1
25-13-13-1
25-25-25-1
25-13-13-13-1
10.72%
10.05%
9.59%
10.01%
9.40%
12.68%
11.86%
12.06%
11.69%
11.66%
11.27%
10.74%
10.44%
10.41%
10.22%
ANN Performance
8%
10%
12%
14%
16%
0 5 10 15 20 25 30 35 40
Epochs
M
A
P
E

Fig. 2. Performance of the 25-13-13-13-1 ANN architecture for the MAPE estimation of the
24
th
hour ahead prediction
2.3 Evaluation of the ANN Output and Confidence Interval Derivation
The output of the ANN is the improved wind power prediction for each studied inter-
val. In order to evaluate the performance of the ANN, the MAPE is calculated com-
paring the outputs of the improved wind power forecast with the actual wind power
production from the wind parks of Crete for the period of study; i.e. 4 last months of
2001. The MAPE index for the ANN is calculated as follows:
Improved Wind Power Forecasting Using a Combined Neuro-fuzzy and ANN Model 109
100% e

=
i
a f
r
P
P P

(1)

=
=
N
i
r
e
N
MAPE
1
1

(2)
where e
r
is the wind power prediction error, P
f
is the forecasted wind power provided
by the ANN, P
a
is the actual wind power, P
i
is the total installed wind power capacity
and N is the number of hours studied (for our application N=24). For the island of
Crete, P
i
=67.35 MW, for the year 2001. Negative values of e
r
mean underestimation
of the ANN output, while positive mean overestimation.
Overestimation of wind power, leads to lack of energy unless sufficient spinning
reserve has been committed to the power system. Thus, the higher the overestimation
of the forecasting model, the higher the spinning reserve that should be maintained,
leading to more units to be committed for the same load and thus the higher the oper-
ating cost. On the other hand, underestimation of wind power has as an impact that
the committed units operate in lower efficiency operating points increasing their oper-
ating cost.
After the evaluation of the improved wind power forecast tool is complete, an 85%
confidence interval of the forecasting error is derived in order to help the operators to
determine their spinning reserve policy as far as wind power forecast uncertainty is
concerned.
3 Crete Power System
Crete is the largest isolated power network in Greece with significant wind power
penetration around 10% of the annual island demand since 2000. The instantaneous
wind power penetration has reached 39% during some valley hours in winter and
early spring [9]. The installed wind power capacity on the island is currently
105.15 MW. There are also installed 690 MW of various thermal units, such as diesel,
gas turbines, steam turbines and one combined cycle unit in three power plants.
Public Power Corporation (PPC) is the operator of this power system and is obliged
to buy at specific price (90% of the retail low voltage price), the energy produced by
the wind park installations. Thus, the improved wind power forecasting and the estima-
tion of its confidence interval are significant, especially during low load periods, when
slow response units, steam turbines and combined cycle units operate to avoid commit-
ting surplus units or not having enough units to compensate for unit loss.
In our study, data from the last 4 months of 2001 was used, when the installed
wind power capacity was 67.35 MW.
4 Results
In Table 2, the MAPE for both the existing neuro-fuzzy wind power forecasting
model and the proposed methodology are presented for the TS data. In all cases, espe-
cially in the 0-10 MW class, the proposed methodology offers much better results.

110 Y.A. Katsigiannis et al.
Table 2. Comparison of TS MAPE for the 0-10 MW class and for the 10-67.35 MW class for
the existing wind power forecasting tool (neuro-fuzzy model) and the proposed model (ANN)
0-10 MW class
MAPE on test test
10-67.35 MW class
MAPE on test test
Estimation
(Hours
ahead) Neuro-fuzzy
model
Improved
model (ANN)
Neuro-fuzzy
model
Improved
model (ANN)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
19.10%
21.70%
20.74%
21.93%
22.27%
21.07%
23.40%
25.58%
20.82%
21.53%
21.46%
24.44%
23.79%
23.14%
23.49%
24.01%
28.10%
24.20%
22.54%
23.67%
23.30%
23.16%
22.40%
24.32%
9.04%
11.88%
11.53%
11.22%
12.32%
11.29%
11.88%
14.56%
11.37%
13.53%
12.10%
14.37%
14.01%
12.48%
11.53%
12.29%
14.57%
12.18%
13.70%
11.73%
13.45%
13.84%
13.12%
12.79%
18.10%
15.38%
16.71%
16.70%
15.41%
15.43%
15.63%
15.36%
16.64%
18.53%
16.94%
17.52%
15.30%
15.45%
16.59%
17.12%
17.68%
17.00%
15.48%
15.53%
17.41%
16.46%
19.44%
17.52%
9.40%
9.54%
9.37%
11.16%
10.03%
10.46%
10.69%
10.43%
11.27%
11.16%
11.02%
11.66%
11.21%
10.82%
11.61%
11.23%
10.00%
9.94%
11.05%
9.51%
10.33%
9.99%
8.96%
10.22%
The MAPE differences range from 8.00% (10
th
hour estimation) to 13.53% (17
th
hour
estimation) for the 0-10 MW class and from 4.09% (13
th
hour estimation) to 10.48%
(23
rd
hour estimation) for the 10-67.35 MW class.
In Tables 3 and 4, the 85% confidence intervals, expressed as 7.5% and 92.5% per-
centiles (ptl) on the test set for the two classes of wind power forecasting values are
presented, according to both the outputs of the existing model and the proposed
method, respectively.
The proposed methodology provides significant reduction to each confidence in-
terval range and much smaller underestimated values, so the power system operator
can estimate the wind power production more accurately avoiding committing more
units than necessary. More specifically, in the 0-10 MW class, the existing models
lowest underestimating errors are always under -40%, while in the proposed model
the corresponding values only once exceed -30%. In the 10-67.35 MW class, the
underestimation error differences are smaller, but in almost every case are over 10%.

Improved Wind Power Forecasting Using a Combined Neuro-fuzzy and ANN Model 111
Table 3. 85% confidence interval (c.i.) for the estimated error of 0-10 MW class and 10-67.35
MW class of the wind power forecasting tool (neuro-fuzzy model)
0-10 MW class 10-67.35 MW class Estimation
(Hours ahead) 7.5% ptl 92.5% ptl 7.5% ptl 92.5% ptl
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
-41.80%
-46.52%
-42.52%
-46.38%
-48.88%
-42.88%
-50.01%
-50.09%
-44.91%
-48.35%
-48.08%
-52.55%
-50.71%
-48.77%
-52.14%
-50.47%
-54.43%
-53.52%
-49.90%
-51.27%
-49.34%
-50.36%
-51.51%
-53.64%
2.04%
0.99%
3.26%
1.56%
2.15%
0.98%
0.44%
0.93%
7.22%
8.37%
8.59%
7.90%
8.08%
2.60%
5.45%
7.77%
0.64%
2.95%
6.19%
7.98%
2.07%
8.47%
3.19%
1.92%
-42.75%
-40.18%
-35.79%
-43.21%
-31.54%
-33.81%
-31.40%
-32.43%
-35.25%
-35.28%
-36.97%
-42.85%
-33.40%
-30.44%
-33.30%
-35.33%
-34.35%
-31.31%
-28.21%
-29.53%
-33.38%
-27.19%
-31.55%
-30.99%
14.80%
15.96%
14.87%
13.47%
20.46%
17.85%
20.88%
13.35%
17.70%
19.20%
21.60%
19.36%
16.28%
26.17%
22.94%
23.81%
25.29%
26.49%
23.93%
24.03%
26.60%
32.02%
27.59%
30.92%

For both classes, smaller differences of the confidence intervals largest overestima-
tion values are observed, especially in the 0-10 MW class, where for some estimations
the initial model gives slightly better results.
Figs. 3 and 4 provide the difference in the forecast and actual operation for specific
values of the studied period for the existing neuro-fuzzy model and the proposed
ANN model. For each case, the minimum and maximum value of the wind power is
also displayed, as it results from the upper and lower boundary of the 85% confidence
interval. The reference date is the 28/12/2001 and prediction time 12:00. This predic-
tion time-series offers wide variation of the predicted values from the neuro-fuzzy
model, the input data of the ANN model, ranging between 5.18 MW and 45.23 MW.
Thus a more representative analysis of models performance can be done. The se-
lected time-series also provides acceptable number of data for both classes of wind
power prediction of the initial model (8 data from the 0-10 MW class and 16 data
from 10-67.35 MW class). The comparison of Figs. 3 and 4 proves that the perform-
ance of the proposed ANN is much better than the existing model, since the ANN
wind power estimation is much more accurate, while its 85% confidence interval is
significantly narrower.
112 Y.A. Katsigiannis et al.
Table 4. 85% confidence interval (c.i.) for the estimated error of 0-10 MW class and 10-67.35
MW class of the proposed ANN
0-10 MW class 10-67.35 MW class Estimation
(Hours ahead) 7.5% ptl 92.5% ptl 7.5% ptl 92.5% ptl
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
-13.12%
-21.88%
-9.50%
-14.83%
-15.91%
-20.91%
-25.95%
-33.77%
-20.40%
-16.93%
-21.77%
-25.00%
-20.07%
-21.21%
-18.18%
-20.30%
-27.83%
-19.05%
-19.19%
-19.20%
-15.55%
-20.42%
-19.44%
-28.20%
1.36%
3.18%
0.70%
1.31%
1.81%
2.83%
2.07%
1.88%
1.87%
4.43%
1.90%
1.54%
3.54%
2.46%
0.80%
1.95%
5.50%
2.62%
2.81%
2.88%
2.32%
5.06%
2.92%
2.69%
-14.94%
-15.56%
-16.73%
-18.17%
-15.72%
-16.32%
-15.90%
-20.45%
-16.48%
-22.03%
-15.97%
-20.39%
-17.29%
-15.86%
-20.65%
-20.03%
-21.10%
-11.62%
-17.35%
-14.00%
-20.10%
-17.02%
-16.32%
-14.99%
6.64%
10.57%
6.25%
4.00%
8.64%
6.84%
6.88%
9.11%
10.23%
6.50%
9.65%
7.32%
9.60%
12.14%
13.39%
9.82%
8.88%
8.42%
10.46%
8.46%
8.01%
8.36%
6.40%
12.34%
0
10
20
30
40
50
60
70
1 3 5 7 9 11 13 15 17 19 21 23
Hours Ahead
W
i
n
d

P
o
w
e
r

(
M
W
)
Forecast
Actual
c.i. Pmin
c.i. Pmax

Fig. 3. Forecast versus actual wind power and min/max boundaries of 85% confidence intervals
for the existing neuro-fuzzy model for the 28/12/2001 at 12:00
Improved Wind Power Forecasting Using a Combined Neuro-fuzzy and ANN Model 113
0
10
20
30
40
50
60
70
1 3 5 7 9 11 13 15 17 19 21 23
Hours Ahead
W
i
n
d

P
o
w
e
r

(
M
W
)
Forecast
Actual
c.i. Pmin
c.i. Pmax

Fig. 4. Forecast versus actual wind power and min/max boundaries of 85% confidence intervals
for the proposed ANN model for the 28/12/2001 at 12:00
The economic impact of the improvements of the wind power forecasting, espe-
cially for confidence interval, is due to the reduction of spinning reserve requirements
to compensate for the wrong estimation of wind. It is considered that the spinning
reserve requirements is given by the following equation:
power wind erval int conf forec load res Spin _ _ _ 1 . 0 _ + = (3)
where conf_interval is the 92.5% percentile (ptl) values used corresponding to the
larger user-defined acceptable wind underestimation level, load_forec is the fore-
casted load and wind_power is the installed wind power capacity.
The impact of reduced spinning reserve is shown in Table 5 for two characteristic
days corresponding to the two classes of the test sets and for different loading condi-
tions. During Day 1, low loading, the wind power forecast never exceeded 10 MW,
while during Day 2, high loading, the wind power forecast was always over 10 MW
so the corresponding data should be used from Tables 3 and 4.
It can be seen that there are significant savings in the operating cost during medium
to high wind power conditions during high load conditions reaching 1.00%. This is
due to the fact that the more expensive gas turbines are committed for less time, or
less of these units are required.
Table 5. Characteristic days used for indicating the impact of improved wind power confidence
interval
Day Total daily
demand
(MWh)
Average Wind
Power Production
(MW)
Characterization Percentage
Savings
Day 1 5197.4 1.3
Low Load, Low Wind
Production
0.05%
Day 2 7396.5 26.6
High Load, Medium
Wind Production
1.00%

114 Y.A. Katsigiannis et al.
5 Conclusions
This paper proposes a combined neuro-fuzzy and ANN model for wind power fore-
casting. The output of an existing neuro-fuzzy wind power prediction tool is used as
input to the proposed ANN structure. It is shown that the proposed ANN model ex-
ploits the past performance of the neuro-fuzzy model and provides more accurate
wind power forecasting values. More specifically, the proposed method offers signifi-
cant improvements in all crucial information for power system operators, concerning
wind power prediction and its uncertainty estimation, providing narrower confidence
intervals for the predicted wind power. Thus the operator can very quickly and very
accurately have improved wind power forecast with narrower confidence intervals
based on the initial wind power forecast provided by the Neuro-fuzzy tool or any
other wind power forecasting algorithm. Thus, the power systems operators have at
their disposal much more accurate information on the expected wind power in the
following few hours, that can be used as inputs for the economic scheduling functions
of the power systems. Reduction of the uncertainty concerning wind power, especially
for autonomous power systems helps in increasing the confidence of the power sys-
tems operators on wind power and its further exploitation.
References
1. Web site of European Wind Energy Association (EWEA): On-line www.ewea.org
2. Hatziargyriou, N., Tsikalakis, A.G., Androutsos, A.: Status of Distributed Generation in
the Greek Islands. Accepted for presentation at the 2006 Power Engineering Society Gen-
eral Meeting of IEEE, Montreal/Canada
3. Nielsen, T.S., Madsen, H.: Statistical Methods for Predicting Wind Power. Proc. of the
European Wind Energy Conference, Dublin Ireland (October 1997) 755-758
4. Wang, X., Sideratos, G., Hatziargyriou, N., Tsoukalas, L.H.: Wind Speed Forecasting for
Power System Operational Planning. 8th PMAPS 2004, Iowa (September 12 16 2004)
5. Kariniotakis, G., Stavrakakis, G.S., Nogaret, E.F.: Wind Power Forecasting Using Ad-
vanced Neural Network Models. IEEE Transactions on Energy Conversion, Vol. 11, No 4.
(December 1996) 762-767
6. Kariniotakis, G.: Forecasting of Wind Parks Production by Dynamic Fuzzy Models with
Optimal Generalisation Capacity. Proc. of the 12th ISAP 03, ISAP 03/032, Lemnos (Sep-
tember 2003)
7. Torres, J.L., Garcia, A., De Blas, M., De Francisco, A.: Forecast of Hourly Average Wind
Speed with ARMA Models in Navarre (Spain). Solar Energy, Vol. 79. (2005) 65-77
8. Giebel, G., Brownsword, R., Kariniotakis, G.: The State of the Art in Short-Term Predic-
tion of Wind Power: A Literature Overview. Deliverable D1.1 of ANEMOS project, avail-
able at: http://anemos.cma.fr
9. Hatziargyriou, N.D., Dimeas, A., Tsikalakis, A., Gigantidou, A., Thalassinakis, E.: Secu-
rity and Economic Impacts of High Wind Power Penetration in Island Systems. 40
th
Cigre
Session, Paris (August 2004)
10. Pinson, P., Kariniotakis, G.N.: Wind Power Forecasting Using Fuzzy Neural Networks
Enhanced with On-line Prediction Risk Assessment. Proc. of 2003 IEEE PowerTech, Bo-
logna, Italy
Improved Wind Power Forecasting Using a Combined Neuro-fuzzy and ANN Model 115
11. Hatziargyriou, N., et al.: MORE CARE Overview. Proc. of MedPower02, MED02/002,
Athens (November 2002)
12. Kariniotakis, G., Mayer, D., Halliday, J.A., Dutton, A.G., Irving, A.D., Brownsword, R.A.,
Dokopoulos, P.S., Alexiadis, M.C.: Load, Wind and Hydro Power Forecasting Functions
of the More-Care EMS System. Proc. of MedPower02
13. Kollias, S.: A Multiresolution Neural Network Approach to Invariant Image Recognition.
Neurocomputing, Vol. 12. (1996) 35-57
14. Doulamis, A., Doulamis, N., Kollias, S.: Retrainable Neural Networks for Image Analysis
and Classification. Proc.of IEEE Int. Conf. on Systems Man and Cybernetics, Orlando,
USA (1997)
15. Kollias, S., Anastassiou, D.: An Adaptive Least Squares Algorithm for the Efficient Train-
ing of Artificial Neural Networks. IEEE Trans. on Circuits & Systems, Vol. 36. (1989)
1092-1101